matmul4_loop.mapal

// matmul4 — the ADR-0021 motivating program: 4×4 matrix multiply as ONE flattened
// loop, building the result cell-by-cell via `c[t] <- v` (array element update).
// The `cell` fn's inner k-loop computes one dot product; the two loops live in
// two fns (the canonical-loop rule is per-fn). Oracle contract for the generated
// arrays below: prints -275 then 3748.
// Perf-scale versions (N=16..128) + cross-language harness: benches/matmul/.

fn cell(a: [f32; 16], b: [f32; 16], i: i32, j: i32) -> f32 {
    mut k: i32   <- 0;
    mut acc: f32 <- 0.0;
    loop {
        (k < 4) -> {
            -true-> {
                acc + a[i * 4 + k] * b[k * 4 + j] -> acc;
                k + 1 -> k;
                -> loop;
            }
            -false-> acc -> ret;
        }
    }
}

fn matmul(a: [f32; 16], b: [f32; 16]) -> [f32; 16] {
    mut c: [f32; 16] <- b;
    mut t: i32       <- 0;
    loop {
        (t < 16) -> {
            -true-> {
                t / 4 -> i;
                t % 4 -> j;
                (a, b, i, j) -> cell -> v;
                c[t] <- v;
                t + 1 -> t;
                -> loop;
            }
            -false-> c -> ret;
        }
    }
}

fn main() {
    [
      -37.0, -30.0, -23.0, -16.0, -9.0, -2.0, 5.0, 12.0,
      19.0, 26.0, 33.0, 40.0, 47.0, -47.0, -40.0, -33.0] -> a;
    [
      7.0, 14.0, 21.0, 28.0, 35.0, 42.0, 49.0, -45.0,
      -38.0, -31.0, -24.0, -17.0, -10.0, -3.0, 4.0, 11.0] -> b;
    (a, b) -> matmul -> c;
    c[0] -> println;
    c[15] -> println;
}